Graph structured skills improve language model task performance
GraphSkillEvo: Evolutionary Optimization of Graph-Structured Agent Skills
Machine Learning
Summary
Large Language Model (LLM) agents can perform better if they have clear instructions, called skills, to guide their actions. The authors found that writing these skills as graphs, where each step is connected and organized, helps the model understand and follow tasks better than just plain text instructions. They created a program called GraphSkillEvo that evolves and improves these graph skills over time, making agents smarter at solving problems. This approach worked better than previous methods on several tests using advanced LLMs.
What this means in practice
- •For ai system developers: Enhance conversational AI agents by using graph-structured skills for clearer task workflows and improved task accuracy.
- •For automation engineers: Create more reliable automated workflows by evolving structured skill graphs that reduce redundancy and improve execution.
Authors
Rui Sun, Zhi Zheng, Zhenkun Wang, Zhichao Lu
Abstract
Skills can improve the performance of Large Language Model (LLM) agents by providing task-specific procedural guidance, while skill optimization further improves their effectiveness through iterative refinement. However, existing skill optimization methods typically represent skills as unstructured natural-language instructions, creating two key challenges: 1) Unstructured skills often lack explicit workflow-level guidance and contain substantial redundancy, making them difficult for LLMs to execute; 2) the vast search space of unconstrained natural-language skills makes skill optimization ineffective. To address these challenges, we propose representing skills as graph-structured natural-language artifacts. In graph-structured skills, each node represents an execution step together with its operational guidance, while directed edges encode context-dependent transitions between steps. Compared to unstructured skills, graph-structured skills can provide clear workflow-level guidance. Moreover, the proposed graph-structured skill can also facilitate skill optimization. Building on this structured representation, we introduce GraphSkillEvo, a population-based evolutionary optimization framework with mutation and crossover operators for graph-structured skills. By maintaining multiple candidate skills and combining effective components, GraphSkillEvo enables broader and more comprehensive exploration of the structured skill space than purely LLM-based iterative self-refinement. Extensive experiments across five agent benchmarks demonstrate that GraphSkillEvo consistently outperforms the strong skill optimization baseline SkillOpt, improving average accuracy by 4.01% on GPT-5.4-nano and 1.76% on GPT-5.4. Our code is available at https://github.com/ruisun7/GraphSkillEvo.